Table 1.

The performance of ChatGPT-4’s initial recommendations and revised recommendations on the Gray Zone cases.

Case ID	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
Distribution of votes for the Gray Zone clinical expert recommendations:
Expert 1	61.54	20	5.56	7.14	71.43	60	40	8	29.41	60	37.5	62.5	0	25	16.67
Expert 2	15.38	26.67	0	57.14	0	40	20	0	52.94	30	25	12.5	100	50	50
Expert 3	0	33.33	55.56	35.71	14.29	0	40	32	5.88	10	25	12.5	0	25	33.33
Expert 4	0	20	38.89 -	0	–	–	20	11.76	–	12.5	12.5	–	–	–
Expert 5	23.08	–	–	–	14.29	–	–	40	–	–	–	–	–	–	–
GPT-4’s self-assessment:
Closest	E3	E2	E1	E1	E4	E2	E1	E3	E3	E1+E2 E3		E1	E2+E3	E2	E2+E3
Favourite	E3	E3	E4	E1	E2	E2	E2	E2	E2	E1+E2 E3		E2	E1	E2	E2
Senior physician’s assessment: Initial recommendation
Correctness	4	4	3	4	4	4	3	2	4	3	4	3	4	3	4
Comprehensi.	3	4	3	2	3	2	4	2	4	4	3	3	2	4	4
Novel aspects	Yes	Yes	No	Yes	No	Yes	Yes	Yes	No	Yes	Yes	Yes	Yes	Yes	Yes
Hallucination	No	No	No	No	No	No	No	Yes	No	Yes	No	No	No	No	No
Revised recommendation
Correctness	4	4	4	4	4	4	4	4	4	4	4	4	4	4	4
Comprehensi.	3	4	3	4	4	4	3	4	4	4	3	4	3	4	4
Novel aspects	Yes	No	No	Yes	No	Yes	No	Yes	No	No	No	No	No	No	Yes
Hallucination	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No

Closest: ChatGPT-4’s initial recommendation is closest to which expert’s recommendation.

Favourite: Which expert’s recommendation is the most proper for the patient.

Comprehensi., Comprehensiveness.